REMOTESENSINGFORAGRICULTURE
Today, a growing number of satellite constellations provide a wealth of real-time earth observation data. How can remote sensing data help design more efficient farming systems?
Introduction
Agriculture is a sector in which the use of remote sensing data is particularly relevant. You may have already seen satellite images showing crop development, as in the example above. This picture is a 32-day composite of the Normalized Difference Vegetation Index (NDVI) calculated from red and infrared reflectances measured by the Landsat 8 satellite. The NDVI provides a fairly accurate picture of plant photosynthetic activity.
One of the use cases that comes to mind when thinking about remote sensing applications in agriculture is the use of satellite imagery to reveal within-field heterogeneity, and thus implement variable rate application to address these differences in crop development (for example, different doses of fertilizer depending on crop needs). But satellite data can also be used on larger scales.
Agricultural land covers a large part of the world’s surface and crops grow in very different soil and climatic conditions, which has a strong influence on yields. Remote sensing data can help characterize these growing conditions. In addition, agricultural plots are cultivated by a large number of growers, making direct data collection difficult. Finally, satellite data analysis may provide a means of remunerating farmers who adopt environmentally-friendly practices.
This is what we will illustrate in the following lines with an example. More specifically, the case study presented here concerns changes in the organic matter content of soils on the Limagne plain (France).
Context of the study area
The Limagne plain is a sedimentary plain of about 200,000 hectares near the city of Clermont-Ferrand (France), at an average elevation of 350 metres. Pedogenesis has produced diversified soil types, ranging from shallow to deep calcareous clay soils (calcisols) to heavy and deep clays (vertisols). In the Eastern part of the plain, calcareous sediments have been altered by the Allier River to give sandy alluvium soils (fluvisols).
As can be seen on the map alongside, which shows elevation data supplied by NASA Shuttle Radar Topography Mission, the plain is surrounded by two mountainous landscapes where the land use is grassland used for breeding dairy or suckling cattle.
The climate is semi-continental. A mountainous relief to the West generates a Foehn effect that greatly reduces rainfall; the average annual precipitation was 573 mm in the 2000–2018 period. For the same period, the mean annual temperature was 12.2°C. The average thermal amplitude is moderate (difference of 16.4°C between the average temperature of the hottest and coldest month) but extreme events can occur, especially during the summer when there are periods of several days with temperatures above 40°C. Periods of drought and heat waves are the main factors limiting crop yields.
Land use evolution in Limagne
While parts of the plain have been cultivated since ancient times, historically, heavier clay soils were kept in pasture. This grassland was drained in the 1960s and 1970s and, together with land consolidation, this led to an almost total disappearance of livestock farming.
The figure below illustrates the evolution of the landscape near Joze. It shows the conversion of grassland into cropland along the Allier river, as well as an increase in the size of the fields.


Today, the dominant land use is arable production of soft winter wheat and grain maize, for human consumption or for seed production.
Evolution of soil organic matter
Changes in farming systems on the Limagne plain have led to substantial modifications in soil characteristics. Based on an analysis of historical soil test data of the territory, covering the period from 1954 to 2019, we found that soil organic matter stocks showed a significant decrease in the three main soil types of the study area.
The figure below shows a simplified version of this analysis, comparing average organic matter content values for two periods (before and after 1990):
There are three main reasons for the decline in soil organic matter in the area:
- the end of livestock manure spreading on the fields;
- conversion of grassland into arable land;
- cereal straw export.
About the last point, the export of straw from the plain to the livestock farms in the surrounding mountains is common, but the transport of manure from livestock farms to arable farms on the plain is much rarer, because, due to logistical constraints, manure imports are hardly feasible over distance of more than 20 km. As a result, organic matter exports are not compensated.
Even if local production diversity has beneficial effects on farming systems, the specialization of farms, or even entire regions, is not unique to this area. In particular, the separation of crop and livestock production is a general trend that is difficult to reverse. We therefore looked at ways of stabilizing or increasing soil organic matter content in arable, stockless farming systems of the Limagne plain.
This is where remote sensing data comes into play.
Assessment of main crop development
The figure below shows NDVI trends for the main crops in the region for the year 2020, based on a combination of Landsat 8 and Sentinel-2 multispectral satellite images.
You may hover the image to identify each curve.
As expected, peak vegetation for winter crops (wheat, rapeseed and barley) tends to occur in the spring, while that for spring crops (corn and sunflower) tends to occur later, in the summer. For these last two crops, the average winter NDVI during the period preceding sowing is relatively low (between 0.2 and 0.3, which corresponds to the value of bare soil). This indicates that intermediate crops (i.e. crops that will be sown to cover the soil between two main crops) are poorly developed in the area.
Such a result is consistent with a previous study we carried out to identify the adoption rate of intermediate cover on a French scale, using multispectral images and the NDVI. At country scale, winter soil cover rate before spring-sown crops was estimated between 37% and 48% for 2018, with strong heterogeneities between French departments. The Puy-de-Dôme department, with the Limagne plain, was one of the departments with the lowest adoption rate. Or cover crops are an important lever for increasing soil organic matter stocks in stockless farming systems. Understanding the determinants of adoption of cover crops is therefore important for increasing organic matter storage.
Drivers of cover crop adoption
To understand these drivers, we now need to look at the crop rotation scale. The graph below shows the NDVI evolution for one of the most common crop rotations in Limagne: soft wheat, maize and soft wheat (mean NDVI for all fields following this rotation). For the same period, from January 2018 to December 2020, the plot also shows the soil moisture, using data provided by the NASA-USDA enhanced SMAP global soil moisture dataset
For the rotation presented above, the longest intercropping period (and therefore the most interesting for sowing a cover crop) is between the first wheat crop and the corn crop. However, the slight increase in NDVI over this period shows that, in the Limagne plain, this practice is underdeveloped. An explanation for this can be found in the comparison with soil moisture, which shows the summer drought during the cover crop establishment period. To overcome this limit to cover crop emergence, cover crops may be established by undersowing in wheat during spring, as already experimented by some farmers in the study area.
The second limit to the establishment of cover crops comes at the end of winter when, conversely, the soil is saturated with water, which limits the possibilities of destroying cover crops. This is particularly true for clay soils which are very wet. Furthermore, at the scale of France, by combining multispectral images with other data sources, we have shown that clay content is the main obstacle to the establishment of cover crops. In Limagne, to get around this limitation, some farmers are experimenting with sowing directly in plant cover, with minimal soil tillage (only on the row, using the strip-till technique for example).
But, despite the inherent limitations of the territory, there is considerable variability in vegetation development between plots sharing the same crop rotation.This can be seen in the graph below, which compares the evolution of NDVI for fields cultivated with corn monocultures. Among the more than 300 fields cultivated with this rotation in Limagne over the study period, the curves compare the NDVI of the 5 plots with the lowest or highest soil cover.
In addition to the average NDVI value for each plot over the study period, the graph below also gives an estimate of soil cover, calculated from the NDVI values according to the formulas given in the bibliography. NDVI is well suited to estimate vegetation ground cover, as there is a linear relationship between the two variables, between bare soil and 100% ground cover. Estimating biomass from multispectral images alone is more difficult, as signal saturation occurs above 100% ground cover.
Monitoring vegetation cover
Following the previous examples, thanks to remote sensing data, it is possible to estimate a rate of vegetation cover per field, and to represent these data on a map, as shown below:
The map shows (i) the significant effect of crop rotation on soil vegetation cover and (ii) the variability that can exist between plots with the same rotation. Such output could help foster good agricultural practices by rewarding farmers who apply those practices. For example, we have previously shown that cover crops are more developed in livestock farming regions, where they can be used as additional fodder resources. On the other hand, in cereal-growing regions, these crops are mainly seen as an additional cost. Specific subsidies, based on the analysis of remote sensing data, could help develop cover crops in arable, stockless farms.